Agentic advertising has spent the past year as a demo: an AI agent negotiating a media buy on stage, a proof of concept wired to a single ad server. That phase is ending. IAB Tech Lab’s release of AAMP 2.3, the third revision to its Agentic Advertising Management Protocol in under four months, shifts the standard from experimental scaffolding to infrastructure built for production traffic, real compliance checks, and multiple competing platforms talking to each other.
From Demo to Deployment
AAMP first shipped as a way to describe, in a shared vocabulary, how an autonomous buying agent could discover inventory, request a price, and close a transaction without a human clicking through an ad server UI. AAMP 2.0 arrived in April and proved the concept was transactable. AAMP 2.3, published July 30, is aimed at a different question: can this run at the scale and reliability an enterprise ad operations team actually requires.
The answer, according to IAB Tech Lab COO and EVP of Product Shailley Singh, rests on infrastructure most marketers never see. The protocol now supports full deployment on Amazon Bedrock AgentCore Runtime, in both REST and MCP modes, with S3-backed inventory adapters and CloudFormation provisioning so a buy-side or sell-side team can stand up an agent environment without building it from scratch. Storage now scales horizontally through SQLite, Redis, or combined Postgres and Redis configurations. IAB Tech Lab says the release carries nearly 5,000 unit tests across its agents, a detail that matters less for what it proves about code quality than for what it signals: this is being built to survive an audit, not a keynote.
Governance Becomes the Product
The more consequential change is what AAMP 2.3 bolts onto the transaction layer itself. The IAB Diligence Platform and SafeGuard Privacy are now integrated directly into the protocol’s Buyer Agent as a vendor approval gate, meaning privacy diligence happens inside the automated workflow rather than as a separate manual step a compliance team runs after the fact. AAMP 2.3 also adds a pricing provenance field designed specifically to prevent a scenario unique to agentic buying: an AI agent hallucinating a CPM that doesn’t exist and completing a transaction against it.
That single addition is a tell for how the industry is thinking about the risk in this technology. Autonomous agents transacting real budget do not fail the way traditional programmatic pipes fail, with a dropped bid request or a rendering error. They fail by confidently generating a wrong number and acting on it. Building a provenance check into the protocol itself, rather than leaving it to each platform to catch downstream, is IAB Tech Lab treating hallucination as a solved-for infrastructure risk rather than a model problem individual vendors patch around.
Interoperability, Not a Walled Garden
AAMP 2.3 adds working integrations for Google Ad Manager reporting and Meta Ads, alongside the existing Amazon Bedrock support, and extends model flexibility to OpenAI, Google Gemini, NVIDIA NIM, Ollama, and Hugging Face-hosted models. On the transport side, MCP with Streamable HTTP becomes the primary mechanism (with legacy SSE retained for compatibility), and OAuth replaces older credential handling for the Streaming Hub and Buyer Cloud components.
The audience side of the protocol matured as well. Agentic Audience Extension, built on OpenRTB, is now positioned as ready for fully programmatic and agentic transactions, with Prebid module support and a contextual enrichment integration from Mixpeek, one of two outside contributors (alongside deal-management software from HyperMindz) whose code IAB Tech Lab folded into the open-source AAMP repository. In under three months, the organization says, the GitHub repo has drawn contributions and use cases from across the industry rather than staying an internal Tech Lab project. That is the detail that separates a real standard from a single vendor’s API: competitors are building on the same protocol instead of shipping incompatible versions of the same idea, a pattern the IAB’s push for a shared video ad measurement vocabulary followed on a slower, less automated track.
What This Means for the Marketing Leader
For a CMO or head of media, none of the CloudFormation or MCP transport detail is the point. The point is that the window for treating agentic buying as an experiment is closing faster than most media plans account for. Platforms are already shipping agent-run auction mechanics into production, as the real-money conversational ad auctions now running on conversational commerce platforms show. A protocol maturing around governance and multi-vendor interoperability at the same time means the agencies and in-house teams that wait for a single dominant platform to standardize this will instead find several already interoperating without them.
The practical exposure is compliance, not just capability. If privacy diligence and pricing provenance are becoming protocol-level features rather than platform-specific bolt-ons, procurement and legal teams evaluating any agentic buying tool should be asking whether it implements AAMP-level checks natively or requires a separate audit layer bolted on after the fact. That distinction will increasingly separate vendors built for enterprise budget from vendors built for a demo.
What to Watch Next
Three signals are worth tracking through the rest of the year: whether Google and Meta extend their new AAMP integrations beyond reporting and buying into full agentic transaction support on their own inventory, whether the pricing provenance field becomes a contractual requirement in agency holding company vendor reviews, and whether the GitHub contribution pattern IAB Tech Lab points to continues to broaden past the current handful of named contributors. A protocol only matters if the ecosystem keeps building on it after the press release fades.
Source: IAB Tech Lab
